The share card decides what readers see
Most readers who meet your article meet a card first: a title, a line of description and an image, as covered in Open Graph tags in WordPress. A disclosure at the end of the article is honest and invisible to all of them. If the use of a tool matters to how a reader should take the piece, the card is where it has to show.
That does not mean every post needs an AI label in its title. Many uses of writing tools, such as spelling checks or a first outline, do not change what the reader is getting. The disclosure policy described in AI-generated content in social feeds decides when you disclose; this guide is about where the line goes once you do.
Where a disclosure reaches readers
| Place | Reaches card readers? | Good for |
|---|---|---|
| Title tag or og:title | Yes, always | Pieces that are substantially generated |
| Meta description or og:description | Yes, on most networks | A short factual line about the tool's role |
| First sentence of the post | Often, as fallback text | Newsletters and posts without custom descriptions |
| Byline area | No, but visible on arrival | The standing policy and a link to it |
| Footnote | No | Detail for readers who want it |
Wording that works in a small space
- Say what the tool did, not that a tool was involved: Draft summarised with a language model and checked by the editor.
- Name the human responsibility: Written by Ana Ruiz; images generated and edited by her.
- Avoid vague badges such as AI-assisted with no detail, which readers cannot interpret.
- Keep it under about 100 characters when it goes into a description, so it is not cut off.
Google's guidance for publishers says generative content is judged on whether it helps readers, and suggests explaining how automation was used where readers would reasonably expect it. YouTube and other platforms ask creators to disclose realistic altered or synthetic media. The EU AI Act adds transparency duties for some generated content. A clear line in the description meets the spirit of all three without turning the card into a legal notice.
AI disclosure on social media: a worked example
Take a newsletter site that publishes 20 posts a month and, under its policy, discloses on the 5 where a language model produced the first draft. The numbers are invented for this example. Each of the 5 gets the same 72-character line at the end of its description: First draft by a language model, rewritten and checked by the editor. The other 15 carry no line, and the policy page explains why.
Sharing the 5 posts to four networks gives 20 cards. In the example, 15 cards show the line in full, 3 show it cut after about 100 characters of description, and 2 show no description at all because that network's card layout shows only a title and image. Moving the line to the start of the description fixes the 3 truncated cards on the next test. For the 2 cards with no description, the only option is the title, which the editor decides is too heavy for a draft-assistance note; the disclosure on those posts lives in the first sentence and the byline instead.
Pitfalls
- A different wording on every post. Readers learn a fixed phrase; they do not learn ten variations.
- Disclosure in a category name or tag. Cards never show them, and readers rarely notice them on the page.
- Putting the line at the end of a long description, where most networks cut it.
- Disclosing tools that did not shape the piece. A line on every post about a spelling checker trains readers to ignore the line that matters.
- Changing the description after sharing and assuming the card updates. It updates only when the network refetches, which can take days or a manual scrape.
Check it worked
Run each disclosed post through the network debuggers, then look at a real share from a logged-out account on a phone. Note where the line appeared and where it was cut. Keep the rows; they show which placement works on which network, and the answer changes when networks redesign their cards. The entries below are examples.
| Page | Network | Time (UTC) | Preview showed | File carried |
|---|---|---|---|---|
| /2026/09/market-roundup/ | 2026-09-16 12:04 | Title, full description with disclosure line, image | n/a | |
| /2026/09/market-roundup/ | 2026-09-16 12:11 | Description cut at about 100 characters; line lost | n/a | |
| /2026/09/market-roundup/ | Mastodon | 2026-09-16 12:19 | Full description, author line | n/a |
| /2026/09/market-roundup/ | X | 2026-09-16 12:27 | Title and image only | n/a |
The same routine, with the image checks added, is set out in testing whether the share card matches the page. Other signals that travel with a card are listed in the provenance section.
Keep authorship visible too
A disclosure answers what the tool did; a byline answers who is responsible. Networks that read author tags can show the writer beside the card, as the guide to authorship tags for shared links explains. Together they tell a reader the one thing a card otherwise hides: that a named person stands behind the post.
Sources
- Google Search Central: guidance on using generative AI content
- Regulation (EU) 2024/1689, the AI Act (EUR-Lex)
- YouTube Help: disclosing altered or synthetic content
Questions
Where should an AI disclosure go on a blog post?
In the description or first sentence when it matters to how the piece is read, because those travel with a share. Keep a fuller note on the page.
Do I need to disclose using a grammar checker?
Usually not. Disclosure matters when a tool shaped the content a reader receives, such as drafting, summarising or generating images.
Will a disclosure in the title hurt clicks?
It can reduce clicks from readers who would have felt misled, which is the point. Honest, specific wording tends to cost less than a correction later.
Does Google penalise disclosed AI content?
Google's published guidance says it rewards helpful content however it is produced and targets content made mainly to manipulate rankings.
How long should an AI disclosure line be?
Under about 100 characters when it sits in a description, and placed at the start rather than the end, because several networks cut descriptions short on the card.
Hannah Voss, Editor. Checks every guide against a working WordPress install and the networks' current documentation. Last reviewed September 2026.
Related reading
- AI-generated content in social feeds: what publishers should doHow AI-generated images and text change what happens when your posts are shared, and the practical steps publishers can take: disclosure, provenance, bylines.7 min read
- Authorship tags for shared links: author, article:author and fediverse:creatorAdd authorship signals to link previews: the author meta tag, article:author, twitter:creator and fediverse:creator, with WordPress code and testing steps.6 min read
- Labelling AI-generated images in link previewsHow to label an AI-generated og:image honestly: captions, alt text, digital source type metadata and what networks show when the card is shared.6 min read
- Adding Open Graph tags in WordPressAdd Open Graph tags to WordPress with an SEO plugin or theme code: the four required tags, image tags, per-post overrides, and how to avoid duplicate tags.7 min read